Agent Computer, explained.
Concepts
What is an Agent Computer?
A persistent cloud computer dedicated to a single agent. Each one is a full Linux environment with its own file system, network, and installed tools. It persists across sessions — stop it, resume it days later, and everything is exactly where you left it. Every task runs inside an Agent Computer.
What is an Agent Executor?
The AI runtime inside each Agent Computer — the brain that drives the agent. Rebyte supports multiple executors: Claude Code, Gemini CLI, Codex, and Rebyte (our built-in executor that needs no API key). Each has different strengths. You choose the best executor for each task.
What is Agent Context?
The data your agents can access. Connect your organization's databases, spreadsheets, Google Drive, cloud storage, and internal knowledge. Agents read from these sources to understand your business — no need to re-explain context every time. Agents also write output back, so results from one agent become input for another.
What are Team Skills?
Reusable capabilities your team builds and attaches to Agent Computers. Encode your company's proprietary workflows, internal tool integrations, and domain-specific knowledge into packages that any agent on your team can use. Skills are composable — combine them to handle multi-step workflows. This is how you turn tribal knowledge into agent capabilities.
Execution
How does a task work?
You describe what you need. Rebyte starts an Agent Computer, runs the executor, and delivers the results. You can close your laptop and come back when it's done.
What is a follow-up?
A follow-up is a new message sent to an already running task. It continues the same conversation — the agent keeps its full context, session, and working state. Use it to add clarifications, change direction, or ask the agent to focus on a specific part.
Can an Agent Computer run multiple tasks?
Yes. One Agent Computer holds one VM, and multiple tasks run inside it sequentially. Each task gets its own session, but they all share the same file system, installed tools, and git state. When one task finishes, the next one picks up where the environment left off.
What is the difference between a new task and a follow-up?
A follow-up continues an existing task — same session, same conversation history. A new task starts a fresh session inside the same Agent Computer. The VM, files, and tools carry over, but the agent starts a new conversation without memory of previous tasks.
Is my Agent Computer always running?
No. Agent Computers automatically pause when idle. You only use resources while a task is actively running. When you need the computer again, it resumes with everything preserved — files, state, installed tools.
Access Control
How do organizations work?
An organization is a group of users. All resources — Agent Computers, Team Skills, Agent Context — are scoped to the organization. A shared resource in one org is invisible to members of another org. Users can belong to multiple organizations.
What are the visibility levels for an Agent Computer?
Three levels. Private: only the creator can see it (default). Shared: all organization members can see it, view all tasks inside it, and send follow-ups. Public: anyone with the link can view it read-only, but cannot send follow-ups or create tasks.
What can a team member do with a shared Agent Computer?
View all tasks and their output, send follow-ups to any running task, and create new tasks inside the same computer. They share the same VM, file system, and git state. They cannot change the computer's settings, visibility, or access list — only the creator can.
Can I share an Agent Computer with specific people instead of the whole org?
Yes. The creator can add individual users to a private Agent Computer's access list. Those users get the same access as org members on a shared computer — they can view tasks and send follow-ups — but the computer stays invisible to everyone else in the org.
Do tasks inside an Agent Computer have their own access control?
No. Tasks, documents, and all resources inside an Agent Computer inherit the computer's visibility. There is no per-task access control. If you can access the Agent Computer, you can access everything inside it.
What are the visibility levels for Team Skills?
Two levels. Private: only the creator and explicitly granted users can see and install it. Public (within org): all organization members can discover and install it. Team Skills never leak outside your organization regardless of visibility level.
Who controls Team Skill access?
The skill creator. Only the creator can change visibility, grant or revoke access for specific users, and publish new versions. Version history is immutable — previous versions can be rolled back to at any time.
How does Agent Context access control work?
Agent Context data sources have the same three visibility levels as Agent Computers: private, shared, and public. Each data source has its own access list managed by its creator. Agents can only read from data sources their computer has access to.
Admin
How does agent isolation work?
Every task runs in its own dedicated Agent Computer with a separate file system, network, and process space. Nothing is shared between agents. Admins set per-computer network policies — whitelist allowed domains, block unauthorized endpoints, prevent data exfiltration. Every action is logged.
Can I control which models and API keys my team uses?
Yes. Admins can approve or block specific models, manage API keys centrally, and track usage and cost per team member. You can also bring your own API keys (BYOK) so agents use your organization's provider accounts directly.
Is there an API?
Yes. The Rebyte API lets you create tasks, send follow-ups, poll for status, and retrieve results programmatically. Integrate agent workflows into your CI/CD pipeline, internal tools, or custom dashboards.
How do I integrate Rebyte with my existing services?
Use the Rebyte API to trigger tasks from your existing systems — CI/CD pipelines, internal dashboards, Slack bots, or any service that can make HTTP requests. You can also use webhooks to get notified when tasks complete, so Rebyte fits into your current workflow without replacing anything.
How does Rebyte scale for a team?
Run dozens or hundreds of Agent Computers in parallel — each with its own executor, context, and skills. Shared workspaces give your team real-time visibility into every running agent. Usage, costs, and permissions are all managed centrally.